[Eeglablist] ICA vs. regression-based artifact correction effects on sensor level phase

[Scott Makeig]

Baris -

Strictly speaking, sensor data do not themselves have phase information ...
since nothing (of interest) originates from the sensors themselves. The
phase information summed in sensor signals (by volume conduction) is
(nearly) all summed source signal information. To the extent ICA is
successful, it will minimize the influence of any source waveform on any
other.

Applying ICA optimally is not guaranteed, of course. The two most important
factors are a) the quality of the ICA algorithm, and b) the amount (and
quality) of the decomposition data.   I suggest you use either Extended
Infomax ICA or (still better) Amica, and that you look at some of our
papers and tutorial to understand better how much data is required for the
number of channels you are decomposing.

Jason Palmer (author of Amica) is working with mutual information functions
to better characterize IC subspaces that retain significant residual mutual
information. This may represent spatial dependency, co-modulation, or
appearance of transient dependencies....

I suggest you study and test the SIFT toolbox of Tim Mullen (for Source
Information Flow Toolbox). It uses ICA to go to the source level and then
models event-related source network dependencies (by any of a palette of
measures, with good graphics!)... Regression methods will remove portions
of all (e.g., frontal) sources that also project to the 'EOG' channel(s).

Scott Makeig


2012/2/16 Baris Demiral <demiral.007 at googlemail.com>

> Hi,
>
> Here are the questions:
>
> a) If we take out artifactual ICs (say, eye blinks), do the final
> sensor data loose their crucial phase information?
> b) If we apply linear regression based algorithms to exclude
> artifacts, will this influence the sensor level phase information?
> c) How do these two methods influence sensor based connectivity analysis?
> d) Which sensor-based connectivity measures are robust against volume
> conduction?
>
> I favor source- and ICA-based multivariate connectivity analyses where
> you really do not need to take out ICs, but work on the components of
> interest.
> But, there are plenty of papers out there reporting only pairwise
> sensor connectivity while ignoring the effects of volume conduction
> and artifact correction.
>
> Thanks,
> Baris
> --
> Ş. Barış Demiral, PhD.
> Department of Psychiatry
> Washington University
> School of Medicine
> 660 S. Euclid Avenue
> Box 8134
> Saint Louis, MO 63110
> Phone: +1 (314) 747 1603
>
> _______________________________________________
> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
> To unsubscribe, send an empty email to
> eeglablist-unsubscribe at sccn.ucsd.edu
> For digest mode, send an email with the subject "set digest mime" to
> eeglablist-request at sccn.ucsd.edu
>



-- 
Scott Makeig, Research Scientist and Director, Swartz Center for
Computational Neuroscience, Institute for Neural Computation; Prof. of
Neurosciences (Adj.), University of California San Diego, La Jolla CA
92093-0559, http://sccn.ucsd.edu/~scott
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://sccn.ucsd.edu/pipermail/eeglablist/attachments/20120221/2f2976dc/attachment.html>